AI coding agents: writing, reviewing and shipping code

Market and trends

4 min read · Chapter 6 of 10

Who holds the power

Labs hold the scarce input and now the fastest-growing agent products, and they use supply. Anthropic cut Windsurf's direct access to Claude with under a week's notice in June 2025, when OpenAI was buying it; one of its co-founders said it would be odd to sell Claude to OpenAI. OpenAI said on August 29, 2026 that it would stop supplying Cursor on November 12 under a change-of-control clause after SpaceX's purchase and wouldn't give Cursor its next model, according to press reports. Anthropic stopped Claude subscriptions from covering outside agent tools in April 2026, as reported.
Labs also need the apps. Anthropic reportedly pledged to keep adding compute for Claude in Cursor after OpenAI left, and a leaked draft of its IPO filing reportedly says two unnamed customers made up about a quarter of its 2025 revenue.
App companies escape the labs in three ways. They build or post-train models: Cursor's Composer, Cognition's SWE-2 (post-trained from Moonshot's open-weight Kimi K3, by Cognition's account) and Microsoft's MAI-Code-1-Flash. They route routine work to cheap open-weight models such as Kimi, GLM and Devstral. Or they sell. In July 2025, within days of each other, Windsurf's leaders went to Google and the rest of Windsurf to Cognition; in 2026 SpaceX bought Cursor, in a deal reported at $60 billion.
The repo host holds the only hard gate. GitHub uses it to host rival agents through Agent HQ, meter them by credits and keep the pull request, while shipping its own model to cut cost. Cursor's answer is Origin.
Clouds turn procurement into leverage: Copilot comes inside Microsoft agreements, Kiro and Claude for Enterprise bill through AWS, Gemini Code Assist through Google Cloud, and a purchase on the cloud's bill can count against money a company has already promised that cloud (how much counts is in the contract). Platforms are also the labs' biggest investors and customers.
Compute owners joined in 2026. SpaceX owns Cursor and reportedly rents its Colossus 1 data center to Anthropic, and Nvidia reportedly agreed a $6 billion licence with Poolside, a coding-model company.
Platform teams inside the customer own the setup scripts, CI and dev environments that decide whether background agents work. I think they're the buyer that matters most, which may be why labs are buying environment companies such as Ona.
Customers have little power on price and some on switching (their own keys, several agents in Agent HQ). Their best levers are model routing and caps.

Tailwinds and headwinds

These are the forces I'd expect everyone in AI coding to watch over the next couple of years, and each one helps some players and hurts others. A couple are one force seen from two sides: cheap models help the app companies earn a margin and pull down the labs' price per token.

Tailwinds:

What's changing	Who gains	Who loses
Companies now pay for agent hours on top of seats, and Menlo Ventures, an investor, estimates companies spent about seven times as much on AI coding tools in 2025 as in 2024.	Anthropic, OpenAI; usage-priced tools (Cursor, Cognition, Kiro); GitHub, through its credits	Buyers' budgets; seat-only plans that absorb heavy users
Agents finish longer tasks, and Microsoft said in July 2026 that one in three pull requests on GitHub involves an agent.	Background agents (Cognition, Copilot's cloud agent, Codex cloud, Cursor's cloud agents, Claude Code on the web, Jules); environment owners; GitHub	Autocomplete-only products; per-seat pricing; reviewers' time
Nine in ten professional developers use an agent at work every week, and daily use keeps rising.	Claude Code and Codex, which gained share fastest in JetBrains' 2026 survey	Copilot and Cursor in share, though their revenue still grew
People who don't code now build apps, and Lovable reportedly reached about $500 million of annualized revenue by June 2026.	Lovable, Replit, Bolt; the labs	App builders' margins while they pay list prices; security teams
More code needs checking than people can read, and review and code security became funded categories in 2026.	CodeRabbit, Graphite (Cursor), Socket, Aikido, Semgrep, GitHub; the labs' own security products	Slower incumbents such as Snyk; standalone reviewers if GitHub or the labs bundle review

Headwinds:

What's changing	Who gains	Who loses
The labs sell their own agents to the same buyers as the app companies they supply, and Anthropic and OpenAI have each pulled model supply from a rival tool, OpenAI's from Cursor taking effect in November 2026.	Anthropic, OpenAI, Google	Cursor, Cognition, app builders paying list prices; customers who standardized on one model
Heavy agent users can cost 10 to 100 times their plan, so vendors ration, re-price and pause signups, as GitHub did in April 2026.	Compute owners; vendors with their own cheaper models (Cursor, Microsoft)	Flat-plan vendors and app builders; heavy users; buyers facing price jumps
Finance teams see bills rise without proof of faster shipping, and 38% of organizations now set usage limits, in Stack Overflow's 2026 survey.	Cheap and routed models; seat-capped plans; usage analytics	Usage-priced vendors at renewal (Anthropic, OpenAI, Cursor, Cognition)
Prompt injection and poisoned packages now steer agents, and EU product rules put the liability on the seller.	Review and security vendors; GitHub's hard gate; vendors with sandboxes and the network off by default	Background agents and app builders in regulated accounts; local agents running on developers' credentials
Free command-line agents and cheap open-weight models do routine work at a fraction of frontier prices.	Buyers; app vendors that route to cheap models (Cursor, Cognition, Microsoft)	Frontier labs' price per token; low-end seats

If I could watch only one thing, it would be spend per engineer at renewals, split by model provider, because the move to usage, the labs' gains and finance's pushback all show up there first.